linux-efi.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Fabian Vogt <fvogt@suse.de>
To: Ard Biesheuvel <ardb@kernel.org>
Cc: linux-efi <linux-efi@vger.kernel.org>,
	"Lee, Chun-Yi" <jlee@suse.com>,
	"Loïc Yhuel" <loic.yhuel@gmail.com>,
	"Matthew Garrett" <matthewgarrett@google.com>,
	"Jarkko Sakkinen" <jarkko.sakkinen@linux.intel.com>
Subject: Re: [PATCH] tpm: verify that it's actually an event log header before parsing
Date: Fri, 12 Jun 2020 13:58:51 +0200	[thread overview]
Message-ID: <6807898.V3K9nM3qbd@linux-e202.suse.de> (raw)
In-Reply-To: <CAMj1kXHErV-tx+8GvvLmb-L0wXq1cVwzY+OEr4PEHDZWrB7Dpg@mail.gmail.com>

Am Freitag, 12. Juni 2020, 13:01:33 CEST schrieb Ard Biesheuvel:
> On Fri, 5 Jun 2020 at 16:22, Fabian Vogt <fvogt@suse.de> wrote:
> >
> > It's possible that the first event in the log is not actually a log
> > header at all, but rather a normal event. This leads to the cast in
> > __calc_tpm2_event_size being an invalid conversion, which means that
> > the values read are effectively garbage. Depending on the first event's
> > contents, this leads either to apparently normal behaviour, a crash or
> > a freeze.
> >
> > While this behaviour of the firmware is not in accordance with the
> > TCG Client EFI Specification, this happens on a Dell Precision 5510
> > with the TPM enabled but hidden from the OS ("TPM On" disabled, state
> > otherwise untouched). The EFI claims that the TPM is present and active
> >  and supports the TCG 2.0 event log format.
> >
> > Fortunately, this can be worked around by simply checking the header
> > of the first event and the event log header signature itself.
> >
> > Commit b4f1874c6216 ("tpm: check event log version before reading final
> > events") addressed a similar issue also found on Dell models.
> >
> > Fixes: 6b0326190205 ("efi: Attempt to get the TCG2 event log in the boot stub")
> > Bugzilla: https://bugzilla.suse.com/show_bug.cgi?id=1165773
> > Signed-off-by: Fabian Vogt <fvogt@suse.de>
> > ---
> >  include/linux/tpm_eventlog.h | 14 +++++++++++++-
> >  1 file changed, 13 insertions(+), 1 deletion(-)
> >
> > diff --git a/include/linux/tpm_eventlog.h b/include/linux/tpm_eventlog.h
> > index 4f8c90c93c29..b913faeedcb5 100644
> > --- a/include/linux/tpm_eventlog.h
> > +++ b/include/linux/tpm_eventlog.h
> > @@ -81,6 +81,8 @@ struct tcg_efi_specid_event_algs {
> >         u16 digest_size;
> >  } __packed;
> >
> > +#define TCG_SPECID_SIG "Spec ID Event03"
> > +
> >  struct tcg_efi_specid_event_head {
> >         u8 signature[16];
> >         u32 platform_class;
> > @@ -171,6 +173,7 @@ static inline int __calc_tpm2_event_size(struct tcg_pcr_event2_head *event,
> >         int i;
> >         int j;
> >         u32 count, event_type;
> > +       const u8 zero_digest[sizeof(event_header->digest)] = {0};
> >
> >         marker = event;
> >         marker_start = marker;
> > @@ -198,10 +201,19 @@ static inline int __calc_tpm2_event_size(struct tcg_pcr_event2_head *event,
> >         count = READ_ONCE(event->count);
> >         event_type = READ_ONCE(event->event_type);
> >
> > +       /* Verify that it's the log header */
> > +       if (READ_ONCE(event_header->pcr_idx) != 0 ||
> > +           READ_ONCE(event_header->event_type) != NO_ACTION ||
> 
> Is the READ_ONCE() necessary here?

I'm actually not sure about that. What's confusing me is that both event and
event_header are pointers returned from EFI calls, but only event is passed
through TPM_MEMREMAP if do_mapping is true. event_header is directly
dereferenced without any mapping in any case. I assume it is an unwritten
precondition that event_header is a pointer to mapped memory already? In that
case READ_ONCE might be unnecessary indeed.

> > +           memcmp(event_header->digest, zero_digest, sizeof(zero_digest))) {
> > +               size = 0;
> > +               goto out;
> > +       }
> > +
> >         efispecid = (struct tcg_efi_specid_event_head *)event_header->event;
> >
> >         /* Check if event is malformed. */
> > -       if (count > efispecid->num_algs) {
> > +       if (memcmp(efispecid->signature, TCG_SPECID_SIG,
> > +                  sizeof(TCG_SPECID_SIG)) || count > efispecid->num_algs) {
> 
> So is it guaranteed that every well formed event starts with TCG_SPECID_SIG?

I assume you mean "event log header" here.
This is defined by the TCG EFI Protocol Spec in 5.3 Event Log Header:

A parser may read the first event [...]. The fields of the event log header are
defined to be PCRIndex of 0, EventType of EV_NO_ACTION, Digest of 20 bytes of
0, and Event content defined as TCG_EfiSpecIDEventStruct. This first event is
the event log header.

Table 3 then says about the first 16 bytes, the signature field:

The null terminated ASCII string “Spec ID Event03”.
SHALL be set to {0x53, 0x70, 0x65, 0x63, 0x20, 0x49, 0x44, 0x20, 0x45, 0x76,
0x65, 0x6e, 0x74, 0x30, 0x33, 0x00}.

It references the TCG PC Client Specific TIS, where it says about Table 5:

The 16 bytes Signature for the TCG_EfiSpecIdEvent as defined in this
specification is the NUL-terminated ASCII string “Spec ID Event03”. Based on
this string an evaluator can determine that this log is formatted as crypto
agile log and that the TCG_EfiSpecIdEvent contains information about
recorded digests.

In the case that the event header indicates that it is the log header, it is
still possible that this header is in a format which is not compatible with
struct tcg_efi_specid_event_head. Verification of this signature makes sure
that it is. The way I read the PC Client Specific TIS, only for this specific
string it is guaranteed that it contains the desired information.

> >                 size = 0;
> >                 goto out;
> >         }
> > --
> > 2.25.1



  reply	other threads:[~2020-06-12 11:58 UTC|newest]

Thread overview: 4+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2020-06-05 14:21 [PATCH] tpm: verify that it's actually an event log header before parsing Fabian Vogt
2020-06-12 11:01 ` Ard Biesheuvel
2020-06-12 11:58   ` Fabian Vogt [this message]
2020-06-12 12:16     ` Fabian Vogt

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=6807898.V3K9nM3qbd@linux-e202.suse.de \
    --to=fvogt@suse.de \
    --cc=ardb@kernel.org \
    --cc=jarkko.sakkinen@linux.intel.com \
    --cc=jlee@suse.com \
    --cc=linux-efi@vger.kernel.org \
    --cc=loic.yhuel@gmail.com \
    --cc=matthewgarrett@google.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).